I saw while in POC phase that the cosine similarity done on dense_vector is not good when compared to ScaNN provided by google.
ScaNN is performing way too good in the same vectors we have stored in elasticsearch. Is there any alternative solution to solve this problem? Can't we integrate the external vector search libraries with elasticsearch ?
I can't use ScaNN as a services as it needs vector in memory to calculate the distance. So I'm out of options here.
Can someone help me on this ?